Frontier Kratocracy
From the AI Research Compass wiki
Frontier Kratocracy: Ring title for extreme closedness scores: frontier capability belongs with the few labs able to build it and responsible for securing it.
Beliefs
Frontier Kratocracy is a ring title earned by openness scores below -0.6, the opposite pole of Open-Source Maximalism, on the right ring outside the main axes. Its holders think frontier capability is better concentrated in a few labs with safety teams, that releasing full training data and recipes is too risky, and that the people who can actually build the most capable systems should decide how they are deployed. “Kratocracy” is rule by the strong, and the name is only half a joke: power here comes from the capacity to build.
Because the title concerns governance rather than method, it can be held by a Language scaler or a World-model builder alike. Its orthodox shade reflects that it defends the current structure of the industry, where the largest models are trained by a handful of well-resourced organisations and accessed through APIs.
Arguments
- Controlled API access allows monitoring, rate limiting and rapid patching of misuse, which is impossible once weights are public.
- Detailed technical reports can share evaluation results and safety findings without releasing the capability itself.
- Concentrated resources enable large, sustained investments in safety research, red teaming and infrastructure that dispersed efforts struggle to fund.
Critiques
- Unreproducible results weaken science: claims about closed models cannot be independently verified.
- Concentrating capability in a few organisations raises obvious concerns about accountability, market power and whose values get encoded.
- Openly released models have repeatedly caught up within months, suggesting the security benefit of closedness may be temporary.
Representative works
Relationships
Friends
How to get it
Earned in the full or pro test when: Openness <−0.6.